Frequency-domain Transformer-based Low-light Image Enhancement Network

Abstract

During the process of capturing images under low-light conditions, images often suffer from issues such as low contrast and high noise due to the limitations of illumination. Recently, with the rapid development of transformers, they have shown significant effectiveness in low-light image enhancement tasks. Therefore, we propose a low-light image enhancement network based on the frequency-domain transformer (FDT). This network consists of a main network and a frequency-domain auxiliary network. In the main network, we use the proposed frequency-domain transformer (FDT) to distinguish between high and low-frequency domains and selectively preserve the required high and low-frequency information. In the frequency-domain auxiliary network, the frequency-domain enhancement module (FDEM) is used to extract features and aggregate with the main network, to achieve the purpose of assisting the integration of information from the main network. Extensive experimental results demonstrate that our proposed method outperforms existing methods to a large extent.